With the development of cloud technology, the amount of data to be transmitted and processed becomes increasingly great, and a bandwidth required by a network device and a user device is increased. How to improve the network bandwidth to meet the requirement of the user device becomes an urgent issue to be solved. Presently, the network bandwidth is improved mainly in the following two manners. In a first manner, a processing capability of a single physical link is improved, for example using a chip with a better processing performance and an optical fiber with a better transmission performance. In a second manner, a load is shared by multiple low-speed physical devices and multiple low-speed links, to improve the overall processing and transmission capability. As compared with the first manner, the second manner has obvious advantages in reliability and extensibility. Thus, the technology using the second manner is widely applied to the present network, which may be referred to as “load sharing (LS)” or “load balancing (LB)”.
According to a load sharing method in the prior art, a link for transmitting a data packet is determined by performing a certain calculation on a part of information of the data packet, for example, performing a remainder operation on a hash value of the part of information divided by the number of actual physical links. However, in this manner, if the number of the actual physical links is not an integral power of 2, the “load sharing” has a great deviation due to the remainder operation, and thus the traffic can not be distributed uniformly. If the traffic is not distributed uniformly, for example, a link reaches a full load, a packet on the link may be lost if the traffic is increased continuously, and thereby influencing a service.